Parameter values of a forest landscape model (MOSAIC) are estimated from a terrain sensitive gap model (FACET) over a large number of terrain types by means of distributed computing. The landscape model is a patch transition semi-Markov model that includes fixed and distributed holding times. States are forest cover types defined using stand composition and structure. Terrain types are defined by combining categories for several environmental factors, determined from the sensitivity of the gap and landscape models to these factors along selected gradients. For each terrain type gap-model output is fed to a program that counts transitions between each pair of states, and estimates the fixed lags and the parameters of the probability density functions of the distributed delays. The gap model and the parameter estimator are executed repetitively for many different steps in the gradients and many terrain types taking advantage of a computer cluster running a distributed launching system. Results are evaluated over the entire set of terrain types using the root mean square error between the gap model and the landscape model for each simulation. The method is illustrated by its application the H.J. Andrews Forest in the Oregon Cascades. Six cover types were defined by species dominance and tree height, whereas 300 terrain types were defined combining three environmental factors. Key transitions determine the variation of the landscape-model parameters over all terrain types. The methodology presented here provides an automated, consistent and conceptually clear procedure for scaling up the tree level ecological detail represented by a gap model, to the level of a patch state-transition model for heterogeneous environmental conditions across the landscape. While the method focuses on particular gap and landscape models (FACET and MOSAIC), this approach could be extended to use other fine-scale and landscape models.